Semiconductor wafer film pasting parameter optimization processing method and system

Abstract

The invention relates to the technical field of semiconductor wafer film pasting, and discloses a semiconductor wafer film pasting parameter optimization processing method and system. The method comprises the steps of training a film pasting quality prediction model based on historical production data; constructing a quality evaluation function, and searching an optimal process parameter combination by using an optimization algorithm under a given target condition by taking the maximization of the comprehensive quality score as a target; screening key parameters from the equipment parameters; performing clustering analysis on the wafer characteristic parameters to generate virtual wafer characteristic parameters; training a parameter mapping model, and establishing a mapping relationship between the equipment parameters and the process parameters; a parameter optimization model is constructed, key parameter correction norm minimization is taken as a target, and performance guarantee and robustness constraint conditions are met; and finally, generating optimized equipment parameters based on the correction, and outputting an actual process parameter combination through a parameter mapping model. The stability of the film pasting quality can be improved.

Description

Technical Field

[0001] This application relates to the field of semiconductor wafer bonding technology, and in particular to a method and system for optimizing semiconductor wafer bonding parameters. Background Technology

[0002] In semiconductor wafer lamination processes, traditional methods typically optimize process parameters based on specific batches of wafers and stable environmental conditions. However, in actual production, wafer materials frequently change (e.g., different suppliers or different batches), environmental conditions fluctuate (e.g., humidity, temperature), and equipment hardware states can drift over long periods. These factors lead to fluctuations in lamination yield, resulting in quality issues such as increased bubbles, decreased uniformity, and insufficient adhesion strength. Whenever such problems occur, traditional practices require downtime and extensive trial-and-error adjustments based on the characteristics of the new wafer and the current environment, relying on experience. This not only leads to low production efficiency but also introduces high quality risks.

[0003] Current technologies lack systematic modeling of the dynamic nonlinear relationship between equipment hardware parameters and process parameters, and fail to effectively identify the key equipment parameters that have the greatest impact on film lamination quality. Furthermore, traditional optimization methods often only address typical operating conditions from historical data, failing to cope with potentially extreme wafer characteristics in the future, leading to a sharp decline in the performance of optimized parameter schemes under unknown conditions. Therefore, there is an urgent need for a parameter optimization method that can adapt to wafer variations, environmental fluctuations, and equipment state drift, and possesses strong robustness to improve the stability and generalization ability of the film lamination process.

[0004] Therefore, the present invention provides a method and system for optimizing semiconductor wafer bonding parameters. Summary of the Invention

[0005] The embodiments in this specification provide the following technical solutions: Step S1: Train and generate a film-applying quality prediction model based on historical production data. The film-applying quality model takes process parameter combinations, environmental parameters and wafer characteristic parameters as input parameters and outputs film-applying quality indicators. Step S2: Construct a quality evaluation function, take the film application quality index as input, output a comprehensive quality score, and use the first optimization algorithm to find the optimal combination of process parameters under the given target conditions, with the goal of maximizing the comprehensive quality score. Step S3: Obtain the equipment parameter analysis dataset. Based on the equipment parameter analysis dataset, select key parameters from multiple equipment parameters. Key parameters refer to equipment parameters that contribute significantly to achieving the optimal combination of process parameters. Step S4: Perform cluster analysis on the wafer feature parameters in the historical production data, and extract the vertex coordinates at the boundary vertices of each generated feature cluster to generate multiple virtual wafer feature parameters. Step S5: Train the parameter mapping model based on historical production data. The parameter mapping model takes the combination of equipment parameters, environmental parameters and wafer characteristic parameters as input and outputs the predicted combination of process parameters. Step S6: Using the correction amount of the key parameter as the optimization variable, construct a parameter optimization model. The parameter optimization model takes minimizing the norm of the correction amount of the key parameter as the optimization objective, and simultaneously satisfies the first constraint condition and the second constraint condition. Solve the parameter optimization model to obtain the optimal solution of the parameter optimization model. The optimal solution represents the correction amount of the key parameter. Step S7: Generate optimized equipment parameters based on the correction amount of key parameters. Input the target environmental parameters, target wafer feature parameters and optimized equipment parameters into the parameter mapping model, and use the output of the parameter mapping model as the actual process parameter combination.

[0006] Compared with the prior art, the beneficial effects of the present invention are at least as follows: The technical solution provided in this application systematically seeks the optimal combination of process parameters under given target conditions through a film-applying quality prediction model and quality evaluation function, avoiding reliance on manual trial and error and significantly improving film-applying consistency and yield. By generating virtual wafer feature parameters and applying constraints during the optimization process, it ensures that the optimized parameter scheme does not degrade in performance when facing extreme or unseen wafer features, effectively preventing batch defects. With the goal of minimizing the norm of key parameter correction, it minimizes equipment adjustment range while ensuring performance and robustness, reducing modification costs and operational complexity. Through the parameter optimization model, it simultaneously satisfies historical performance guarantees and future risk control constraints, achieving an optimal balance between performance, robustness, and modification costs, and systematically solving the problem of process instability. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a schematic diagram of an embodiment of a semiconductor wafer film bonding parameter optimization method in this application. Figure 2 This is a schematic diagram of one embodiment of a semiconductor wafer film bonding parameter optimization processing system according to the present application. Detailed Implementation

[0009] This application provides a method and system for optimizing semiconductor wafer lamination parameters. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0010] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the semiconductor wafer film bonding parameter optimization method in this application includes: Step S1: Train and generate a film-applying quality prediction model based on historical production data. The film-applying quality model takes process parameter combinations, environmental parameters and wafer characteristic parameters as input parameters and outputs film-applying quality indicators. Specifically, the following problems exist during wafer lamination: When laminating wafers from Company A, the lamination effect is excellent, with few bubbles and high uniformity. However, when using wafers from a different company or different batches of the same company, the lamination quality becomes unstable. For example, there are significantly more bubbles. Furthermore, during the rainy season, increased humidity in the workshop slightly reduces the yield even when laminating the same type of wafer. The traditional approach is to stop production each time a problem occurs, readjusting parameters based on experience according to the characteristics of the new wafer and the current environment, involving extensive trial and error. This leads to low production efficiency and high quality risk. Therefore, we provide a semiconductor wafer lamination parameter optimization method. Through data and algorithms, this method not only finds the optimal combination of process parameters and optimal equipment parameters but also proactively prevents the aforementioned problems from occurring.

[0011] First, historical production data is collected to establish an ideal target, namely the optimal combination of process parameters. To obtain the optimal combination of process parameters, a film application quality prediction model is first generated using machine learning algorithms. To train an accurate film application quality prediction model, historical production data is collected. Historical production data refers to relevant data from past film application production, including combinations of process parameters, environmental parameters, wafer characteristic parameters, and corresponding film application quality indicators. Combinations of process parameters include film application speed, film application pressure, and film application temperature; environmental parameters include temperature, humidity, and film application batch; wafer characteristic parameters include wafer warpage, wafer surface roughness, and wafer thickness; and film application quality indicators include the number of air bubbles, film application uniformity, and adhesion strength. The combination of process parameters, environmental parameters, and wafer characteristic parameters are used as input parameters, and the film application quality indicators are used as output parameters.

[0012] Step S2: Construct a quality evaluation function, take the film application quality index as input, output a comprehensive quality score, and use the first optimization algorithm to find the optimal combination of process parameters under the given target conditions, with the goal of maximizing the comprehensive quality score. Specifically, to provide optimization targets for subsequent equipment parameter optimization, a quality evaluation function is first constructed. Film application quality has multiple indicators (number of bubbles, uniformity, adhesive strength, etc.), which may conflict with each other. For example, increasing pressure to improve adhesive strength may lead to more bubbles. Therefore, a quality evaluation function is created to integrate multiple film application quality indicators into a single comprehensive quality score. For instance, weights of -0.5, 0.3, and 0.2 are assigned to the number of bubbles, uniformity, and adhesive strength distribution, respectively. These weights reflect process preferences; in this example, reducing bubbles is considered most important, so it is given the highest absolute weight. The quality evaluation function is a mathematical formula: Comprehensive Quality Score = -0.5 * Number of Bubbles + 0.3 * Film Application Uniformity + 0.2 * Adhesive Strength.

[0013] Next, an optimization scenario (i.e., fixed target conditions) is set. The target conditions include fixed environmental parameters and fixed wafer characteristic parameters. For example, the optimization scenario is set to 50% humidity, 25 degrees Celsius temperature and B123 batch film application. The wafer characteristic parameters are the wafer characteristic parameters corresponding to wafers produced in the same batch. In this embodiment, it is assumed that the wafer characteristic parameters of the same batch are the same.

[0014] Finally, under the given objective conditions, with the goal of maximizing the overall quality score, the first optimization algorithm is used to find the optimal combination of process parameters. The first optimization algorithm can be a genetic algorithm or a particle swarm optimization algorithm. The specific method for finding the optimal combination of process parameters will be explained in detail later.

[0015] This step involves using fixed target conditions, such as the wafer characteristic parameters and environmental parameters that we are about to apply the film to, and finding the theoretically optimal combination of process parameters through the film application quality prediction model and the first optimization algorithm.

[0016] Step S3: Obtain the equipment parameter analysis dataset. Based on the equipment parameter analysis dataset, select key parameters from multiple equipment parameters. Key parameters refer to equipment parameters that contribute significantly to achieving the optimal combination of process parameters. Specifically, film application equipment has many hardware parameters, such as the hardness of the rollers and the tension of the film feeding mechanism. However, not all hardware parameters have an equal impact on the film application quality. According to the Pareto principle, only a few parameters, such as 20% of the equipment parameters, often determine the majority of the results. To identify the critical few parameters, limited resources should be invested in these key parameters to achieve the highest cost-effectiveness. There may be complex coupling relationships between equipment parameters. Optimizing all parameters simultaneously not only increases the computational load exponentially but also easily leads to getting trapped in local optima. Therefore, focusing on key parameters can greatly simplify the problem. Based on the above reasons, key parameters are selected from multiple equipment parameters. The specific selection method will be explained in detail later. Key parameters refer to equipment parameters that contribute significantly to achieving the optimal combination of process parameters. Optimizing key parameters can bring about significant performance improvements.

[0017] Step S4: Perform cluster analysis on the wafer feature parameters in the historical production data, and extract the vertex coordinates at the boundary vertices of each generated feature cluster to generate multiple virtual wafer feature parameters. Specifically, there are differences between wafers. Even for the same type of wafer product, there are differences between different batches. Moreover, there are always wafers with significant problems in various characteristics. In future film lamination production, if new combinations of process parameters and new equipment parameters are found, but a batch of wafers that we have never seen in the historical production process is encountered, it may lead to a large number of defects. Therefore, in order to anticipate the most abnormal situation that may be encountered in the future, we perform cluster analysis on the wafer feature parameters in the historical production data. At the boundary vertices of each generated feature cluster, we extract the vertex coordinates at the vertex to generate multiple virtual wafer feature parameters.

[0018] By generating multiple virtual wafer feature parameters, each representing a combination of features of the wafer under the most demanding conditions, the subsequent optimization process is not only improved from fitting known data to ensuring against unknown risks, but also introduces unseen virtual samples on the boundary of the feature space into the subsequent optimization constraints. This forces the optimization process to consider extreme cases and effectively prevents the optimized parameter scheme from performing well only in the vicinity of historical data points, while experiencing a sharp decline in performance when dealing with real wafers in unexplored areas of the feature space.

[0019] Step S5: Train the parameter mapping model based on historical production data. The parameter mapping model takes the combination of equipment parameters, environmental parameters and wafer characteristic parameters as input and outputs the predicted combination of process parameters. Specifically, during wafer lamination, there is a dynamic and non-linear dependency between process parameter combinations and equipment parameters. For example, if a softer roller is used, the optimal pressure value required to achieve the same lamination effect must be changed. To accurately capture the dependency between the two, a parameter mapping model is trained using machine learning methods. The parameter mapping model can predict the corresponding process parameter combinations given equipment parameters, environmental parameters, and wafer characteristic parameters. Through the parameter mapping model, the difficult-to-quantify equipment parameters are linked to the process parameter combinations, enabling the system to predict the expected process behavior under any given equipment, environment, and product configuration, providing data support for subsequent optimization and decision-making.

[0020] Step S6: Using the correction amount of the key parameter as the optimization variable, construct a parameter optimization model. The parameter optimization model takes minimizing the norm of the correction amount of the key parameter as the optimization objective, and simultaneously satisfies the first constraint condition and the second constraint condition. Solve the parameter optimization model to obtain the optimal solution of the parameter optimization model. The optimal solution represents the correction amount of the key parameter. Specifically, in semiconductor wafer lamination processes, traditional methods typically optimize process parameters based on specific batches of wafers and stable environments. When faced with changes in wafer materials (such as different suppliers or batches), fluctuations in environmental conditions (such as humidity changes), and long-term drift in equipment hardware status, these methods exhibit unstable performance and insufficient generalization ability, leading to fluctuations in lamination yield and frequent downtime for debugging.

[0021] To address the aforementioned issues, a constrained optimization problem is constructed and solved to determine the optimal correction amounts for a set of key parameters. These optimal correction amounts minimize the magnitude of the changes under two constraints.

[0022] First, the correction amounts of the key parameters are formalized as optimization variables and initialized, transforming the engineering problem into a computable mathematical problem. Then, a parameter optimization model is constructed, with the optimization objective being to minimize the norm of the correction amounts of the key parameters. To ensure the performance benchmark of the optimization scheme when dealing with historical cases, first and second constraints are applied.

[0023] The first constraint is: the difference between the predicted process parameter combination obtained by inputting the optimized equipment parameters, target environment parameters and historical wafer characteristic parameters into the parameter mapping model and the optimal process parameter combination is less than a first preset threshold.

[0024] The optimized equipment parameters refer to the new key parameters formed by adding corrections to the key parameters, and then combining the key parameters with other non-key parameters of the equipment parameters. The target environmental parameters are the same as the fixed environmental parameters in the target conditions in step S2, which can be the environmental parameters of the upcoming film-coating production. The historical wafer characteristic parameters refer to all typical wafer characteristic parameters that have appeared in history. By inputting the optimized equipment parameters, target environmental parameters, and historical wafer characteristic parameters into the parameter mapping model, the most suitable combination of predicted process parameters based on historical experience can be obtained. The combination of predicted process parameters is compared with the optimal combination of process parameters obtained in step S2, and the difference between the two is constrained to be less than a first preset threshold. By applying the first constraint condition, the performance benchmark of the optimization scheme in processing historical typical wafer products is ensured, performance regression is prevented, and the optimization direction is not sacrificed for the processing quality of historical typical wafer types in pursuit of other goals.

[0025] The second constraint is as follows: For each virtual wafer feature parameter, the optimized equipment parameters, target environment parameters, and virtual wafer feature parameters are input into the parameter mapping model to obtain the first predicted process parameter combination. The first predicted process parameter combination, target environment parameters, and virtual wafer features are input into the film bonding quality model to obtain the first predicted comprehensive quality index. The first comprehensive quality index is input into the quality evaluation function to obtain the first comprehensive quality score. The comprehensive quality score is greater than or equal to the second comprehensive quality score, which is obtained based on the equipment parameters before optimization.

[0026] To ensure that the optimized hardware parameters do not degrade in performance when faced with process variations and extreme operating conditions, a second constraint is applied. For each virtual wafer feature parameter, the optimized equipment parameters, target environment parameters, and virtual wafer feature parameters are first input into a parameter mapping model to obtain a first predicted process parameter combination. Then, the first predicted process parameter combination, target environment parameters, and virtual wafer features are input into a film-attaching quality model to obtain a first predicted comprehensive quality index. Based on a quality evaluation function, a first comprehensive quality score is obtained. The second constraint requires that the first comprehensive quality score be greater than the second comprehensive quality score, which is obtained based on the equipment parameters before optimization.

[0027] The second comprehensive quality score is calculated as follows: the equipment parameters before optimization, the target environment parameters, and the virtual wafer feature parameters are input into the parameter mapping model to obtain the second predicted process parameter combination. The equipment parameters before optimization refer to the initial equipment parameters before the addition of corrections. The second predicted process parameter combination, the target environment parameters, and the virtual wafer features are input into the film bonding quality model to obtain the second predicted comprehensive quality index. The second comprehensive quality index is input into the quality evaluation function to obtain the second comprehensive quality score.

[0028] The parameter optimization model with the first and second constraints added is input into the quadratic programming solver for numerical calculation. An efficient global search is performed to obtain the optimal solution that satisfies all conditions, and the amount of modification of the key parameters that can be executed and quantified is obtained.

[0029] The above method improves process capability by optimizing key parameters, achieving a lasting and universally applicable effect. In a rigorous data model, it balances three key dimensions: performance, robustness, and modification cost, and outputs the optimal solution with the best overall cost-effectiveness. In the process of finding the optimal solution, the worst case is also transformed into a computable and reusable explicit constraint. The system solves the process instability problem caused by uncertainty in the film coating process of semiconductor wafers, and outputs a device parameter modification scheme that achieves the optimal balance in terms of performance, robustness, and cost.

[0030] Step S7: Generate optimized equipment parameters based on the correction amount of key parameters. Input the target environmental parameters, target wafer feature parameters and optimized equipment parameters into the parameter mapping model, and use the output of the parameter mapping model as the actual process parameter combination.

[0031] Specifically, when applying a film to the wafer, the correction amount of the key parameters obtained in step S6 is applied to the key equipment parameters to obtain the corrected key parameters. The corrected key parameters and non-key parameters are combined to generate optimized equipment parameters. The optimized equipment parameters, target environmental parameters, and target wafer characteristic parameters are input into the parameter mapping model to obtain the corresponding process parameter combination. The target environmental parameters and target wafer characteristic parameters are the same as the target conditions in step S2. The target environmental parameters are the actual environmental parameters when applying the film, and the target wafer characteristic parameters are the wafer product characteristic parameters when applying the film. The optimized equipment parameters are applied to the corresponding film application equipment, and the process parameter combination output by the parameter mapping model is applied to the actual process configuration.

[0032] In summary, this application first trains a film-attachment quality prediction model based on historical data. Using this model, under given target conditions, a first optimization algorithm searches for the theoretically optimal combination of process parameters. Then, through contribution analysis, it selects the key parameters that contribute most to achieving the optimal combination of process parameters from numerous equipment parameters. Next, a parameter mapping model is constructed as a mapper connecting hardware devices and process parameters. Using this parameter mapping model as a bridge, a parameter optimization model is established to solve for the optimal correction amount of the key parameters. The optimization must satisfy a first constraint (performance guarantee) to ensure that the processing capability for wafer film attachment after optimization is close to the optimal combination of process parameters, and a second constraint (robustness guarantee) to ensure that the processing capability for various virtual extreme wafers after optimization is not lower than the existing level. Finally, the optimized hardware configuration is combined with environmental parameters and wafer characteristic parameters to generate the most reasonable combination of actual process parameters through the parameter mapping model.

[0033] In one specific embodiment, the selection of multiple key parameters that contribute significantly to achieving the optimal combination of process parameters includes the following steps: Production data before and after equipment hardware changes are extracted from historical production data. The production data is correlated with equipment parameters to form an equipment parameter analysis dataset. Equipment parameters are used as explanatory variables, and each process parameter in the corresponding combination of process parameters in the production data is used as the target variable. Multiple multiple regression models are constructed, and the multiple multiple regression models are fitted based on the equipment parameter analysis dataset. The standardized regression coefficient of each equipment parameter with respect to each target variable is calculated. For each equipment parameter, the absolute values ​​of its regression coefficients in each multiple regression model are added to obtain the comprehensive contribution of the equipment parameter. The top N equipment parameters with the highest comprehensive contribution are selected as key parameters.

[0034] Specifically, to select key parameters from a large number of equipment parameters, the production data before and after equipment hardware changes are first extracted from historical production data. This production data reflects the changes before and after the hardware changes. The production data is then correlated with the corresponding equipment parameters to obtain an equipment parameter analysis dataset. The equipment parameters in this dataset are used as explanatory variables, and the process parameters in the corresponding production data are used as target variables. Multiple multiple regression models are constructed, and each model is fitted to the equipment parameter analysis dataset. The standardized regression coefficient for each equipment parameter with respect to each target variable (i.e., each process parameter) is calculated. The standardized regression coefficient eliminates the influence of dimensions, and its absolute value directly reflects the degree of influence of the equipment parameter on the process parameter. For a single equipment parameter, it will have a regression coefficient in multiple multiple regression models. The sum of the absolute values ​​of these multiple regression coefficients is used as the overall contribution of the equipment parameter. A high overall contribution indicates that changing this equipment parameter will have a significant impact on multiple process parameters. Key parameters are selected based on the overall contribution. If N key parameters are to be selected, the top N equipment parameters with the highest overall contribution are chosen as the key parameters.

[0035] In one specific embodiment, generating multiple virtual wafer feature parameters specifically includes the following steps: The historical wafer feature parameters are clustered using a clustering algorithm to form multiple feature clusters. For each feature cluster, its convex hull in the corresponding feature space is calculated, the vertex coordinates of the convex hull are obtained, and the vertex coordinate values ​​are used as virtual wafer feature parameters.

[0036] Specifically, in order to generate representative virtual wafer samples for testing and ensuring the robustness of the process scheme in subsequent optimization steps, it should be noted that the virtual wafer samples are not real wafers, but represent extreme cases that may occur for various types of wafers in the existing historical data.

[0037] Based on the similarity of a large number of historical wafer feature parameters, clustering algorithms (such as K-means) are used to divide the wafer feature parameters into multiple feature clusters. Different feature clusters may come from different suppliers or use different base materials, and the optimal coating parameters corresponding to each feature cluster may be completely different. Therefore, it is necessary to generate exclusive extreme samples for each type of wafer. In order to generate extreme samples, for each feature cluster, its convex hull in the corresponding feature space is calculated. The convex hull is the minimum convex boundary formed by all data points in the feature cluster. This minimum convex boundary can cover all possible feature combinations of the wafer of the corresponding feature cluster. After calculating the convex hull, the vertex coordinates of the convex hull are obtained. The vertex coordinates are a set of feature parameter values. Each vertex represents a virtual wafer that reaches the limit on one or more features in the wafer of the corresponding feature cluster. Therefore, the vertex coordinate values ​​are used as virtual wafer feature parameters. For example, for a feature cluster, one vertex of its convex hull might be: [warpage = 0.25mm, roughness = 0.035μm, thickness = 780μm, ...], representing an extremely warped and extremely rough virtual wafer.

[0038] In one specific embodiment, the first optimization algorithm is used to find the optimal combination of process parameters, which specifically includes the following steps: The process parameter combinations are obtained from historical production data, and then multiple new process parameter combinations that are different from the historical process parameter combinations are generated. The historical process parameter combinations and the generated new process parameter combinations are combined with the given target conditions to generate multiple input parameters for the film-applying quality model. The input parameters are input into the film-applying quality model to obtain the corresponding film-applying quality index. The film-applying quality index is then input into the quality evaluation function to obtain the corresponding comprehensive quality score. Based on the comprehensive quality score, some process parameter combinations are eliminated, and the remaining process parameter combinations are optimized using the first optimization algorithm to obtain the optimal process parameter combination.

[0039] Specifically, to obtain the optimal combination of process parameters, multiple combinations of process parameters from historical production data are first acquired. To broaden the search scope, several new combinations of process parameters different from the historical combinations are generated. The historical and new combinations are used as the search set. Each combination of process parameters is combined with given target conditions to generate multiple input parameters for the film coating quality model. The target conditions include fixed environmental parameters and fixed wafer feature parameters. The input parameters are input into the film coating quality model to obtain the corresponding film coating quality index. The film coating quality index is then input into the quality evaluation function to obtain the corresponding comprehensive quality score. Based on the comprehensive quality score, some process parameter combinations are eliminated, for example, the half of the process parameter combinations with the lowest film coating quality scores are deleted. For the remaining process parameter combinations, the first optimization algorithm is used to optimize them to obtain the optimal combination of process parameters. The first optimization algorithm can use a genetic algorithm or a particle swarm optimization algorithm, etc. Through the above method, under a given optimization scenario, based on the predictive ability of the film coating quality model, the process parameter combination with the highest comprehensive quality score can be obtained through an intelligent search method (genetic algorithm or particle swarm optimization algorithm).

[0040] In one specific embodiment, training and generating a film-mounted film quality prediction model includes the following steps: Based on the combination of process parameters, the historical production data is clustered at the first level to obtain multiple first datasets. A first label is defined for each first dataset. The first dataset is then clustered at the second level to obtain multiple second datasets. A second label is defined for each second dataset. A pool of lightweight models with different predefined structures is used to generate models and compute key features for each second dataset. For each second dataset, train all models in the model pool sequentially, evaluate the performance of each model, and label the best performing model as the preferred model for the corresponding second label; The classification model is trained by taking key features as input and the corresponding preferred model type as output.

[0041] Specifically, to improve the prediction accuracy of the film-applied coating quality prediction model, multi-level clustering is first performed on the historical production dataset. The clustering method can be K-means clustering. First, coarse-grained clustering is performed based on the combination of process parameters to obtain multiple first datasets. A first label is defined for each first dataset, representing the process mode of the corresponding first dataset. Then, a second-level clustering is performed on each first dataset to obtain multiple second datasets. The second-level clustering can be based on environmental state parameters and wafer feature parameters. A corresponding second label is defined for each second dataset, representing the process scenario of the second dataset. Then, a model pool is generated by predefining multiple lightweight models with different structures, such as linear regression models, multinomial regression models, and support vector regression models. These models are used as candidate models. For each second dataset, corresponding key features are calculated. Key features are used to describe the inherent characteristics of the data under the corresponding process scenario, including: the number and density of data points, the variance and skewness of the quality indicators, and the statistics of the correlation coefficient matrix between process parameters and quality indicators.

[0042] Then, for each process scenario (second label), all models in the model pool are trained once using the corresponding second dataset, and the performance of each model is evaluated. Specifically, the model performance can be evaluated by cross-validation to calculate the mean squared error of each model, and the model with the best performance is marked as the preferred model for the corresponding process scenario.

[0043] Finally, the key features of all the second datasets are used as input, and the corresponding preferred model type is used as the output label to train the classification model, which is used to recommend the model most likely to perform well for new and unlearned process scenarios.

[0044] Whenever it is necessary to predict the screen protector quality index, the screen protector prediction model first analyzes the second label of the input data, calculates the key features of the corresponding second dataset based on the second label, inputs the key features into the classification model to obtain the corresponding preferred model, instantiates the corresponding type of model from the model pool as the quality prediction model, and inputs the input data into the corresponding quality prediction model to obtain the corresponding screen protector quality index.

[0045] The above describes a method for optimizing semiconductor wafer bonding parameters in the embodiments of this application. The following describes a system for optimizing semiconductor wafer bonding parameters in the embodiments of this application. Please refer to [link to relevant documentation]. Figure 2 One embodiment of the semiconductor wafer film bonding parameter optimization processing system in this application includes: The model training unit trains and generates a film-attached quality prediction model based on historical production data. The film-attached quality model takes process parameter combinations, environmental parameters, and wafer characteristic parameters as input parameters and outputs film-attached quality indicators. The target optimization unit constructs a quality evaluation function, takes the film application quality index as input, and outputs a comprehensive quality score. With the goal of maximizing the comprehensive quality score, the first optimization algorithm is used to find the optimal combination of process parameters under the given target conditions. The parameter filtering unit acquires the equipment parameter analysis dataset and filters key parameters from multiple equipment parameters based on the equipment parameter analysis dataset. Key parameters refer to equipment parameters that contribute significantly to achieving the optimal combination of process parameters. The sample generation unit performs cluster analysis on the wafer feature parameters in historical production data, and extracts the vertex coordinates at the boundary vertices of each generated feature cluster to generate multiple virtual wafer feature parameters. The mapping training unit then trains a parameter mapping model based on historical production data. The parameter mapping model takes equipment parameter combinations, environmental parameters, and wafer characteristic parameters as inputs and outputs predicted process parameter combinations. The parameter optimization unit uses the correction amount of the key parameters as the optimization variable to construct a parameter optimization model. The parameter optimization model takes minimizing the norm of the correction amount of the key parameters as the optimization objective, and simultaneously satisfies the first constraint condition and the second constraint condition. Solving the parameter optimization model yields the optimal solution of the parameter optimization model, which represents the correction amount of the key parameters. The parameter generation unit generates optimized equipment parameters based on the correction amount of key parameters. It inputs the target environmental parameters, target wafer feature parameters, and optimized equipment parameters into the parameter mapping model, and uses the output of the parameter mapping model as the actual process parameter combination.

[0046] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0047] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0048] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for optimizing semiconductor wafer film bonding parameters, characterized in that, The method includes: Step S1: Train and generate a film-applying quality prediction model based on historical production data. The film-applying quality model takes process parameter combinations, environmental parameters and wafer characteristic parameters as input parameters and outputs film-applying quality indicators. Step S2: Construct a quality evaluation function, take the film application quality index as input, output a comprehensive quality score, and use the first optimization algorithm to find the optimal combination of process parameters under the given target conditions, with the goal of maximizing the comprehensive quality score. Step S3: Obtain the equipment parameter analysis dataset. Based on the equipment parameter analysis dataset, select key parameters from multiple equipment parameters. Key parameters refer to equipment parameters that contribute significantly to achieving the optimal combination of process parameters. Step S4: Perform cluster analysis on the wafer feature parameters in the historical production data, and extract the vertex coordinates at the boundary vertices of each generated feature cluster to generate multiple virtual wafer feature parameters. Step S5: Train the parameter mapping model based on historical production data. The parameter mapping model takes the combination of equipment parameters, environmental parameters and wafer characteristic parameters as input and outputs the predicted combination of process parameters. Step S6: Using the correction amount of the key parameter as the optimization variable, construct a parameter optimization model. The parameter optimization model takes minimizing the norm of the correction amount of the key parameter as the optimization objective, and simultaneously satisfies the first constraint condition and the second constraint condition. Solve the parameter optimization model to obtain the optimal solution of the parameter optimization model. The optimal solution represents the correction amount of the key parameter. Step S7: Generate optimized equipment parameters based on the correction amount of key parameters. Input the target environmental parameters, target wafer feature parameters and optimized equipment parameters into the parameter mapping model, and use the output of the parameter mapping model as the actual process parameter combination.

2. The method according to claim 1, characterized in that, The first constraint is: The difference between the predicted process parameter combination obtained by inputting the optimized equipment parameters, target environment parameters, and historical wafer characteristic parameters into the parameter mapping model and the optimal process parameter combination is less than a first preset threshold.

3. The method according to claim 1, characterized in that, The second constraint is: For each virtual wafer feature parameter, the optimized equipment parameters, target environment parameters, and virtual wafer feature parameters are input into the parameter mapping model to obtain the first predicted process parameter combination. The first predicted process parameter combination, target environment parameters, and virtual wafer features are input into the film bonding quality model to obtain the first predicted comprehensive quality index. The first comprehensive quality index is input into the quality evaluation function to obtain the first comprehensive quality score. The comprehensive quality score is greater than or equal to the second comprehensive quality score, which is obtained based on the equipment parameters before optimization.

4. The method according to claim 1, characterized in that, Several key parameters that contribute significantly to achieving the optimal combination of process parameters were selected, including: Production data before and after equipment hardware changes are extracted from historical production data. The production data is correlated with equipment parameters to form an equipment parameter analysis dataset. Equipment parameters are used as explanatory variables, and each process parameter in the corresponding combination of process parameters in the production data is used as the target variable. Multiple multiple regression models are constructed, and the multiple multiple regression models are fitted based on the equipment parameter analysis dataset. The standardized regression coefficient of each equipment parameter with respect to each target variable is calculated. For each equipment parameter, the absolute values ​​of its regression coefficients in each multiple regression model are added to obtain the comprehensive contribution of the equipment parameter. The top N equipment parameters with the highest comprehensive contribution are selected as key parameters.

5. The method according to claim 1, characterized in that, Generate multiple virtual wafer feature parameters, including: The historical wafer feature parameters are clustered using a clustering algorithm to form multiple feature clusters. For each feature cluster, its convex hull in the corresponding feature space is calculated, the vertex coordinates of the convex hull are obtained, and the vertex coordinate values ​​are used as virtual wafer feature parameters.

6. The method according to claim 1, characterized in that, The first optimization algorithm is used to find the optimal combination of process parameters, including: The process parameter combinations are obtained from historical production data, and then multiple new process parameter combinations that are different from the historical process parameter combinations are generated. The historical process parameter combinations and the generated new process parameter combinations are combined with the given target conditions to generate multiple input parameters for the film-applying quality model. The input parameters are input into the film-applying quality model to obtain the corresponding film-applying quality index. The film-applying quality index is then input into the quality evaluation function to obtain the corresponding comprehensive quality score. Based on the comprehensive quality score, some process parameter combinations are eliminated, and the remaining process parameter combinations are optimized using the first optimization algorithm to obtain the optimal process parameter combination.

7. The method according to claim 1, characterized in that, Training and generating a film application quality prediction model includes: Based on the combination of process parameters, the historical production data is clustered at the first level to obtain multiple first datasets. A first label is defined for each first dataset. The first dataset is then clustered at the second level to obtain multiple second datasets. A second label is defined for each second dataset. A pool of lightweight models with different predefined structures is used to generate models and compute key features for each second dataset. For each second dataset, train all models in the model pool sequentially, evaluate the performance of each model, and label the best performing model as the preferred model for the corresponding second label; The classification model is trained by taking key features as input and the corresponding preferred model type as output.

8. A semiconductor wafer film bonding parameter optimization processing system, used to implement the semiconductor wafer film bonding parameter optimization processing method as described in any one of claims 1-7, characterized in that, The system includes: The model training unit trains and generates a film-attached quality prediction model based on historical production data. The film-attached quality model takes process parameter combinations, environmental parameters, and wafer characteristic parameters as input parameters and outputs film-attached quality indicators. The target optimization unit constructs a quality evaluation function, takes the film application quality index as input, and outputs a comprehensive quality score. With the goal of maximizing the comprehensive quality score, the first optimization algorithm is used to find the optimal combination of process parameters under the given target conditions. The parameter filtering unit acquires the equipment parameter analysis dataset and filters key parameters from multiple equipment parameters based on the equipment parameter analysis dataset. Key parameters refer to equipment parameters that contribute significantly to achieving the optimal combination of process parameters. The sample generation unit performs cluster analysis on the wafer feature parameters in historical production data, and extracts the vertex coordinates at the boundary vertices of each generated feature cluster to generate multiple virtual wafer feature parameters. The mapping training unit then trains a parameter mapping model based on historical production data. The parameter mapping model takes equipment parameter combinations, environmental parameters, and wafer characteristic parameters as inputs and outputs predicted process parameter combinations. The parameter optimization unit uses the correction amount of the key parameters as the optimization variable to construct a parameter optimization model. The parameter optimization model takes minimizing the norm of the correction amount of the key parameters as the optimization objective, and simultaneously satisfies the first constraint condition and the second constraint condition. Solving the parameter optimization model yields the optimal solution of the parameter optimization model, which represents the correction amount of the key parameters. The parameter generation unit generates optimized equipment parameters based on the correction amount of key parameters. It inputs the target environmental parameters, target wafer feature parameters, and optimized equipment parameters into the parameter mapping model, and uses the output of the parameter mapping model as the actual process parameter combination.

Images